[Obs] 4.4 - Readout error mitigation

[mpharrigan]

Add a facility to take readout calibrations by measuring the observables under the empty circuit with readout symmetrization turned on. This can be plumbed through so other observables measured with measure_observables can be corrected. Note that this all only works for symmetric readout error enforced through readout_symmetrization.

Part of #3647

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[mpharrigan]

Do we have an asymmetric readout channel available in non-contrib cirq? I'd like to keep that in my test

[95-martin-orion]

Do we have an asymmetric readout channel available in non-contrib cirq? I'd like to keep that in my test

Meaning a noise model like DepolarizingWithDampedReadoutNoiseModel? AFAIK all of those are either in contrib or in cirq_google (e.g. the per-qubit version). I wouldn't object to moving them to cirq-core, though.

I'm not sure what "asymmetric readout" means here, but you may be able to get something similar either with from_noise_model_like or by creating your own noise model:

# Simple model that adds amplitude damping after each moment
model_1 = cirq.NoiseModel.from_noise_model_like(cirq.amplitude_damp(0.1))

# More complex model that adds bitflips before measurement
class ReadoutErrorModel(cirq.NoiseModel):
    def noisy_operation(self, op):
        if isinstance(op.gate, MeasurementGate):
            return [cirq.bit_flip(0.1), op]
        return op

[mpharrigan]

asymmetric: p00 != p11

[95-martin-orion]

asymmetric: p00 != p11

In that case, the model_1 example above would fit your requirements, although it may require some finagling to fit your test. (Specifically, the noise is added after every moment, instead of before readout)

[mpharrigan]

vendored a noise model in the test, ptal